Power without control
Agents, with their ability to act, decide and move across systems independently, promise a step change in productivity. But the same autonomy that drives value is creating new forms of cyber risk, challenging organisations' ability to maintain control when things go wrong.
We set out to answer three critical questions:
- How is agentic AI reshaping the cyber threat landscape?
- Where are the gaps in governance, visibility and response that hinder safe deployment at scale?
- What defines cyber resilience when failure is inevitable?
What 800+ senior leaders told us
To answer these questions, Economist Enterprise, supported by Rubrik, surveyed more than 800 business decision-makers globally, all deploying AI agents in live environments, and interviewed leading experts to explore how organisations are managing risk as autonomy scales.
As agents scale, exposure multiplies
The market for AI agents is forecast to grow 44-fold over the next decade. As agents spread across systems, workflows and data environments, they create more pathways for disruption to operations, information and decision-making.
Failure is inevitable.
98% of organisations have already experienced a disruptive agent-related incident.
Nine in ten say more are coming, regardless of the safeguards in place.
The threat is no longer at the perimeter
Cybersecurity was built to keep attackers out. But AI agents are shifting risk inside the organisation, where unauthorised actions, operational blind spots and agent-to-agent interactions can trigger disruption at scale.
Awareness is high, readiness is not
Leaders recognise that cyber risk is changing. But many organisations still lack the visibility, response and testing capabilities needed to operate safely at scale.
When agents fail, the business pays
The consequences of agent-related incidents extend far beyond IT. They can disrupt operations, erode trust and create financial, legal and reputational costs.
The illusion of preparedness
95% of organisations have set mean time to recover targets. But for many, those targets remain informal or loosely defined, creating a false sense of readiness when incidents occur.
- Response/Recovery
- Prevention
Security spending hasn’t caught up with reality
Cybersecurity investment still favours prevention over response and recovery, and leaders expect this imbalance to persist. Yet as risk moves inside the organisation, fortifying the walls is no substitute for fixing the foundations.
Boards are flying blind
A quarter of organisations regularly report cyber-recovery performance to senior leadership, equipping them with the visibility needed for robust decision-making. But for the remainder, board visibility remains out of reach.
Three capabilities define cyber resilience
You cannot control what you cannot see. Yet nearly two-thirds of organisations lack full visibility into their AI agents, making it harder to detect, contain and reverse failures when they occur.
- Not yet assessed whether we can undo or roll back harmful agent actions
- Incomplete or untested capabilities to undo harmful agent actions
- Defined and functioning processes to reverse most, but not all harmful agent actions
- Robust, fully-tested capabilities that reliably reverse harmful agent actions
When incidents occur, speed determines impact. Yet seven in ten organisations lack robust, tested capabilities to contain, reverse and recover from harmful agent actions.
More than a quarter of organisations have yet to define the minimum systems and operations needed to survive a major incident. Of those that have, only 40% test them regularly.
To learn more about how control defines success in the agentic age, read the white paper.
Case studies from Rubrik
How Zurich US built AI-ready resilience across a multi-cloud enterprise
As its AI capabilities expanded, the organisation sought a new level of reversibility
How EōS Fitness is balancing AI risk and innovation
Stonger recovery capabilities are helping the chain of gyms unlock more benefits of agentic AI benefits